22
$\begingroup$

If $S \times \Bbb{R}^k$ is homeomorphic to $T \times \Bbb{R}^k$ for some $k \geq 1$ and $S$ is compact, can we conclude that $T$ is compact?

The case where $k=1$ and $S$ and hence also $T$ are locally connected is dealt with in my answer to this question. I am now curious about the general case.

$\endgroup$
  • 2
    $\begingroup$ The choice of $S$ and $T$ is unfortunate. I first read them as $S$ for the circle and $T$ for the torus. Are they just random spaces? $\endgroup$ – Fly by Night May 8 '13 at 15:40
  • 2
    $\begingroup$ Yes, they are arbitrary. $\endgroup$ – Alexander Thumm May 8 '13 at 15:51
  • 1
    $\begingroup$ It feels to me like some kind of cohomology at infinity calculation could show this, but I don't actually see an argument. $\endgroup$ – Cheerful Parsnip May 8 '13 at 17:41
  • 1
    $\begingroup$ I can't even think of two non-homeomorphic bases such that the cylinders are homeomorphic... The only thing that is obvious to me is that connectedness of the basis is invariant. Nice question. $\endgroup$ – Julien May 8 '13 at 17:45
  • 5
    $\begingroup$ @julien: Bing's dogbone space and the Whitehead manifold are both spaces which become homeomorphic to $\mathbb R^4$ when you cross with $\mathbb R$. The first is not a manifold and the second one is a $3$-manifold. Such things happen a lot in higher dimensions. $\endgroup$ – Cheerful Parsnip May 8 '13 at 18:07
7
+100
$\begingroup$

Yes, if $S$ is compact, then $T$ is also compact.

In fact, the property that $S$ is compact can be expressed in terms of the topological space $X=S\times\mathbb{R}^ k$. The idea is natural enough, if $S$ is compact then it is possible to remove a compact set from $X$ which stops certain maps from the $(k-1)$-sphere $\mathbb{S}^{k-1}$ from being null-homotopic. The precise statement is a little tricky though.

The following are equivalent.

  1. $S$ is compact.
  2. There exists a compact $A\subseteq X$ such that, for every compact $B\supseteq A$ there exists a compact $C\supseteq B$ such that, for every $x\in X\setminus C$, there exists a continuous $f\colon \mathbb{S}^{k-1}\to X\setminus B$ containing $x$ in its image and which is null-homotopic as a map to $X$ but not null-homotopic as a map to $X\setminus B$.

First, suppose that 1 is true. We need to construct the sets $A$, $C$ and the function $f$ for any choices of $B$ and $x$. Set $A=S\times\lbrace0\rbrace$. Given any compact $B\supseteq A$, let $\pi_{\mathbb{R}^k}(B)$ be its projection onto $\mathbb{R}^k$. As this is compact, it is contained in the closed ball $\bar B_R(0)$ for some $R > 0$. Set $C=S\times\bar B_R(0)$. Then, given any $x\in X\setminus C$, we can write $x=(s_0,y_0)$ where $\lVert y_0\rVert = r > R$. Define $f\colon \mathbb{S}^{k-1}\to X$ by $f(a)=(s_0,ra)$. This is null-homotopic as a map to $X$ via the homotopy $F(a,u)=(s_0,(1-u)ra)$. On the other hand, it is not null-homotopic as a map to $X\setminus B$. To see this, define $g\colon X\setminus B\to \mathbb{S}^{k-1}$ by $g(s,y)=y/\lVert y\rVert$. Then, $g\circ f$ is the identity on $\mathbb{S}^{k-1}$. However, it is standard that the (k-1)-sphere is not contractible or, equivalently, the identity on $\mathbb{S}^{k-1}$ is not null-homotopic, implying that $f$ is not null-homotopic.

Conversely, suppose that $S$ is not compact. We need to prove that 2 is false. To be explicit, the negation of 2 can be written as

2'. For any compact $A\subseteq X$ there exists a compact $B\supseteq A$ such that, for any compact $C\supseteq B$, there exists an $x\in X\setminus C$ such that every continuous $f\colon\mathbb{S}^{k-1}\to X\setminus B$ with $x$ in its image and which is null-homotopic as a map to $X$ is also null-homotopic as a map to $X\setminus B$.

So, we need to construct the set $B$ and point $x\in X\setminus C$ for any choice of $A$ and $C$, and show that any such function $f$ is null-homotopic as a map to $X\setminus B$.

Given a compact $A\subseteq X$, its projection onto $\mathbb{R}^k$ is compact, so contained in the closed ball $\bar B_R(0)$ for some $R > 0$. Also, its projection $S_1=\pi_S(A)$ is compact. Take $B=S_1\times\bar B_R(0)$. If $C$ is a compact set containing $B$ then the projection $\pi_S(C)$ is compact, so is not the whole of $S$. Therefore, we can choose $s_0\in S\setminus\pi_S(C)$. Set $x=(s_0,0)\in X\setminus C$.

Now, suppose that $f\colon\mathbb{S}^{k-1}\to X\setminus B$ has $x$ in its image and is null-homotopic as a map to $X$. We need to construct a null-homotopy for $f$ as a map to $X\setminus B$. Write $f(a)=(f_1(a),f_2(a))$ where $f_1,f_2$ are maps from $\mathbb{S}^{k-1}$ to $S$ and $\mathbb{R}^k$ respectively. First, there will exists continuous maps $f^\prime_2\colon\mathbb{S}^{k-1}\to\mathbb{R}^k\setminus\lbrace0\rbrace$ which are arbitrarily close to $f_2$. One way to see this is to first approximate $f_2$ by a smooth function. Then, as the image of a smooth map from $\mathbb{S}^{k-1}$ to $\mathbb{R}^k$ cannot contain any open ball about the origin (its image must have zero measure), there will be arbitrarily small $v\in\mathbb{R}^k$ such that $f^\prime_2=f_2+v$ does not pass through the origin. Then, $f$ will be homotopic to $(f_1,f_2^\prime)$ for $f_2^\prime$ close enough to $f_2$. Next, by scaling, $f$ is homotopic to $(f_1,rf^\prime_2)$ for any $r\ge1$. So, by scaling $f_2^\prime$, we can assume that $\lVert f^\prime_2\rVert > R$.

We have shown that $f$ is homotopic to $(f_1,f^\prime_2)$ where $f^\prime_2\colon\mathbb{S}^{k-1}\to\mathbb{R}^{k-1}\setminus\bar B_R(0)$. Next, by projection onto $S$, the condition that $f$ is null-homotopic as a map to $X$ implies that $f_1$ is null-homotopic. Also, as $f$ has $(s_0,0)$ in its image, $f_1$ has $s_0$ in its image so is homotopic to the constant map to $s_0$. This gives a homotopy of $f$ to the map $a\mapsto(s_0,f^\prime_2(a))$. Finally, by scaling, $f^\prime_2$ is homotopic to the constant map to $0$. So, we have constructed a null-homotopy for $f$ (as a map to $X\setminus B$) to the constant map $a\mapsto(s_0,0)$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.